See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/376780920

Artificial intelligence in rehabilitation: comparing OpenAI ChatGPT

recommendations to evidence-based meta-analyses

Article in Physiotherapy Review · January 2023

DOI: 10.5114/phr.2023.133714

CITATIONS READS

0 366

2 authors:

Sebastian Rutkowski Alejandro Galán-Mercant

Opole University of Technology Universidad de Cádiz
72 PUBLICATIONS 1,243 CITATIONS 63 PUBLICATIONS 1,022 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Sebastian Rutkowski on 03 January 2024.

The user has requested enhancement of the downloaded file.

 Review, Physiotherapy Review, 2023, 27(4), 5-16

Artificial intelligence in rehabilitation: comparing

OpenAI ChatGPT recommendations to
evidence-based meta-analyses

Sebastian Mateusz Rutkowski1 , Alejandro Galán-Mercant2,3

1 Faculty of Physical Education and Physiotherapy, Opole University of Technology, Opole, Poland
2 INIBICA, Institute of Biomedicine of Cádiz, Spain
3 MOVE-IT Research Group, Department of Nursing and Physiotherapy, Faculty of Health Sciences, University of
Cádiz, Spain

Correspondence to: Sebastian Mateusz Rutkowski, email: [email protected]

DOI: https://doi.org/10.5114/phr.2023.133714
Received: 23.03.2023 Reviewed: 04.04.2023 Accepted: 06.04.2023

Abstract
Background: ChatGPT, short for Chat Generative against the existing evidence on rehabilitation
Pre-Trained Transformer, is a language model de- interventions from systematic reviews and me-
veloped by OpenAI that is capable of generating ta-analyses in the Cochrane Library database.
human-like text responses to various prompts. In
Results: The findings of the study indicate that
the field of rehabilitation, ChatGPT has the po-
the AI's recommendations align with the current
tential to be a valuable resource for patients and
scientific knowledge on rehabilitation interven-
clinicians alike. In light of the open, easy access
tions, as demonstrated through a comparison
to this tool, its application in rehabilitation should
with the Cochrane Library database.
be considered and evaluated.
Conclusions: Using AI language models like
Aims: This study aimed to test the feasibility of
ChatGPT can help identify effective rehabilitation
using the OpenAI ChatGPT tool to generate clin-
methods for common conditions.
ical recommendations in rehabilitation and com-
pare its findings with recent meta-analyses.
Material and methods: The selection of rehabili-
tation areas was based on a report on the global
estimates of the need for rehabilitation, with the
Key words
top three disorders being musculoskeletal disor- rehabilitation, Artificial Intelligence, AI,
ders, neurological disorders, and chronic respira- ChatGPT.
tory diseases. The authors evaluated the respons-
es of ChatGPT to clinical questions in these areas

5
Physiotherapy Review | Volume XXVII Issue 4/2023

Introduction
Within two months of its release, ChatGPT has OpenAI's tool, ChatGPT, has demonstrated ver-
exerted significant influence on e-commerce and satility in multiple domains, such as writing mo-
disrupted the education and entertainment in- tivational letters for job applicants, generating
dustries [1]. ChatGPT, short for Chat Generative high-quality recipes based on available ingredi-
Pre-Trained Transformer, is a language model ents, and creating investment strategies based on
developed by OpenAI that is capable of generating the user's asset structure. It has also been tested
human-like text responses to various prompts. by scientists from Axios, who reported that the
This model is based on the Transformer archi- answers provided by the algorithm were consist-
tecture, which was introduced in 2017 by Vaswani ent, comprehensive, and in-depth, passing the
et al. [2] and has since become the dominant ap- United States Medical Licensing Examination
proach in Natural Language Processing (NLP). The (USMLE) [7]. Moreover, Springer Nature has tak-
Transformer architecture is designed to handle en a clear stance on using AI-based tools in pre-
sequential data, such as text, using self-attention paring scientific works. Neither ChatGPT nor any
mechanisms to weigh the importance of differ- other similar tool can be signed as the authors of
ent input parts. This allows the model to capture works that are published by the publisher. How-
long-range dependencies between words, which
ever, the publisher still needs to completely cut
is crucial for tasks such as language translation,
off from this technology and allow its use as long
where the meaning of a word can depend on its
as the authors properly inform about the use of
context within a sentence [3].
AI-based tools in the writing process [8]. Fur-
ChatGPT is a variant of the Transformer architec- thermore, the Chris Stokel-Walker article identi-
ture that has been pre-trained on a large corpus fied four articles in which ChatGPT was shown
of text data from the internet, including websites, as a co-author, two of them being indexed in the
books, and social media. Once pre-trained, the PubMed resources. Therefore, it is likely that uti-
model can be fine-tuned for a wide variety of NLP lization of this kind of instrument will experience
tasks, including text classification, question an- a substantial rise.
swering, and text generation. In the case of text
In the field of rehabilitation, ChatGPT has the
generation, ChatGPT can generate coherent and
potential to be a valuable resource for patients
fluent text responses to a given prompt, making
it useful for conversational agents, chatbots, and and clinicians alike. It can provide patients with
other applications that require natural language a convenient and accessible way to engage in re-
interaction [4]. ChatGPT boasts a noteworthy habilitation exercises and receive feedback on
capability of producing real-time, high-quality their progress. For clinicians, it can be used to
text outputs, rendering it a useful tool for appli- automate the delivery of rehabilitation programs,
cations requiring human-like communication [5]. reducing the time and effort required to create
However, it was rapidly established that this tool and manage these programs. However, while the
could be extremely effective in completing home- potential benefits of ChatGPT in rehabilitation
work, taking exams, and even writing complex are significant, it is important to note that its use
essays (1). As a result of this, access to ChatGPT in this context must be carefully considered and
was blocked in New York public schools in Janu- evaluated. As with any technology, some poten-
ary, and similar bans have been reported in other tial risks and limitations must be taken into ac-
schools across the United States. However, con- count, such as privacy concerns, the potential for
sidering that a simple mobile phone is enough to incorrect or biased responses, and the need to
access the full capabilities of the AI tool, these ensure the quality and validity of the rehabilita-
bans are symbolic in nature [6]. tion programs being delivered.

6
 Physiotherapy Review | Volume XXVII Issue 4/2023

In light of the open, easy access to this tool, its of the available evidence, considering the quali-
application in the field of rehabilitation should ty and limitations of individual studies [9]. It was
be considered and evaluated. Furthermore, an presumed that searching for answers to a clin-
examination of the PubMed scientific reposito- ical question about treatment effectiveness in
ry revealed a mere five publications that investi- the Cochrane Library could provide a high-level
gate the prospective utilization of ChatGPT in the overview of the current state of knowledge and
medical domain; however, none of these works increase confidence in the conclusions drawn
pertain to the subject of rehabilitation. As a re- from the available evidence. The responses were
sult, the purpose of this narrative review was to examined on February 3rd and 4th, 2023.
test the feasibility of using the OpenAI ChatGPT
tool to generate clinical recommendations in the Selection of conditions
rehabilitation field and compare its findings with The selection of rehabilitation areas was based on
recent meta-analyses in various rehabilitation a report, "Global estimates of the need for reha-
areas. The structure of this narrative review is bilitation based on the Global Burden of Disease
organized as follows: The second section out- Study 2019: a systematic analysis for the Global
lines the electronic search resources employed. Burden of Disease Study 2019", published in the
The Results section is divided into three distinct Lancet (10). The results indicated that in 2019, 2.41
subsections, each highlighting the top three most billion individuals worldwide had conditions that
frequently rehabilitated conditions. The ensu- would benefit from rehabilitation, leading to 310
ing portion features a discussion of the findings, million years of life lived with disability. The find-
while the final section offers conclusive remarks. ings of this study challenge the commonly held
view that rehabilitation services are required only
Electronic search by a small portion of the population and highlight
It was decided that the process of comparing the need for rehabilitation to be integrated into
answers to clinical questions will be presented primary healthcare and made accessible to com-
according to the following diagram (Fig. 1). The munities globally. Therefore, it was decided to
Cochrane Library database is a well-established prioritize the top three disorders: musculoskel-
and respected source for systematic reviews etal disorders (2412 million cases), neurological
(SR) and meta-analyses on the effectiveness of disorders (255 million cases), and chronic res-
healthcare interventions. These types of studies piratory diseases (118 million cases).
provide a comprehensive and rigorous synthesis

Figure 1. Study
flowchart.
Developing a question Comparison
for ChatGPT of responses

Identify the most Cochrane database

Conclusions
common disorders analysis

7
Physiotherapy Review | Volume XXVII Issue 4/2023

Results
Musculoskeletal disorders Verification of responses to the Cochrane Data-
The report found that musculoskeletal disorders base's Systematic Reviews
contribute the most to the need for rehabilitation It was decided to evaluate the responses against
services. Approximately 1.71 billion people have the existing evidence on rehabilitation inter-
musculoskeletal conditions, with low back pain ventions. One of the most recently founded me-
being the most prevalent and leading health con- ta-analyses was a study by Hayden et al. [11],
dition contributing to the need for rehabilitation which aimed to evaluate the effect of exercise
services in 134 of the 204 countries analyzed. therapy on pain and functional limitations in
adults with chronic non-specific low back pain
ChatGPT question:
in comparison to control conditions such as no
“What are the effective treatments for chronic
treatment, standard care, placebo, and other
lower back pain?”
conservative treatments. Authors found moder-
ChatGPT answer: ate-certainty evidence that exercise therapy is
"The most effective treatments for chronic lower probably effective for treating chronic low back
back pain in comparison to no treatment, usual pain compared to no treatment, usual care, or
care, or placebo can vary depending on the indi- placebo, as demonstrated by small but signifi-
vidual's specific condition and symptoms. How- cant improvements in pain and functional limi-
ever, some commonly used and evidence-based tations outcomes. A study evaluating the effec-
treatments for chronic lower back pain include: tiveness of acupuncture for chronic non-specific
low back pain was also found [12]. Mu et al. [13]
• Exercise therapy: This can include physical
noticed low-certainty evidence that acupuncture
therapy, strengthening exercises, and aerobic
may provide short-term pain relief compared to
exercises.
a sham intervention, but the difference did not
• Medications: Over-the-counter pain relievers meet a clinically important threshold. The results
such as ibuprofen and acetaminophen, as well of the analysis indicate that adverse events were
as prescription pain medications, can be effec- similar in the acupuncture and sham intervention
tive for managing chronic lower back pain. groups and included injection site pain, bruising,
• Spinal manipulation: This may include chi- and worsening of lower back pain. The Cochrane
ropractic care, osteopathic manipulation, or resource also includes a study evaluating the ef-
massage therapy. fectiveness of spinal manipulative therapy for
• Acupuncture: This traditional Chinese med- chronic low-back pain. Rubinstein et al. found
icine technique involves the insertion of fine high-quality evidence that spinal manual therapy
needles into the skin at specific points to re- had a small, statistically significant but not clini-
lieve pain. cally relevant, short-term effect on pain relief and
functional status compared to other interven-
• Interventional procedures: This can include
tions. The effect was seen when SMT was added
nerve blocks, epidural injections, and spinal
to another intervention as well, but the quality of
cord stimulation.
evidence in these cases varied from low to high.
It is important to consult with a healthcare pro-
A study by Wieland et al. [14] revealed low-cer-
fessional to determine the most appropriate and
tainty evidence from 11 trials showed a small im-
effective treatment plan for an individual's spe-
provement in back-specific function with yoga,
cific condition and symptoms.”

8
 Physiotherapy Review | Volume XXVII Issue 4/2023

and moderate-certainty evidence from nine trials pain. French et al. [17] found limited evidence to
showed a small improvement in pain compared support the use of superficial heat and cold for
to no exercise at three months. Wiliams et al. low back pain. The evidence showed that heat
[15] reported that Cognitive Behavioral Therapy wrap therapy provided a small short-term reduc-
(CBT) has small or very small beneficial effects for tion in pain and disability for a population with
reducing pain, disability, and distress in chronic a mix of acute and sub-acute low-back pain, and
pain, while Schaafsma et al. [16] not found CBT to the addition of exercise further reduced pain and
be conclusively superior to physical conditioning, improved function. However, there is insufficient
either as an alternative or as an adjunctive thera- evidence to evaluate the effects of cold for low-
py. Also, a systematic review was found on the ef- back pain, and the evidence for the application of
fectiveness of superficial heat or cold for low back cold treatment is even more limited (Table 1).

Table 1. Verification of responses regarding musculoskeletal disorders.

ChatGPT Cochrane Database

Exercise therapy Confirmed

Manual therapy Confirmed

Cognitive-behavioral therapy Not confirmed

Mind-body interventions Confirmed

Heat and cold therapy Not confirmed

Acupuncture Confirmed

Neurological disorders • Physical therapy: to improve mobility, balance,

and coordination through exercises and move-
Global Burden of Disease Study 2019 revealed
ment training.
that, among neurological disorders, stroke rep-
resented the greatest need for rehabilitation (86 • Occupational therapy: to help with activities
million people). of daily living such as dressing, eating, and
grooming.
ChatGPT question:
• Speech therapy: to improve communication
“What are the effective methods of rehabilitation
and swallowing abilities.
after stroke?”
• Cognitive rehabilitation: to improve thinking,
ChatGPT answer: memory, and problem-solving skills.
“Rehabilitation after stroke typically involves • Aquatic therapy: to take advantage of the buoy-
a combination of physical, occupational, and ancy and resistance of water for rehabilitation
speech therapy. Effective methods of rehabilita- exercises.
tion may include:

9
Physiotherapy Review | Volume XXVII Issue 4/2023

• Electrical stimulation: to improve muscle idence of a small improvement in basic activities

strength and coordination using electrical of daily living immediately after the intervention
stimulation. (Functional Independence Measure) and at the
three- and six-month follow-ups.
It is important to note that the type and inten-
sity of rehabilitation will vary depending on the Moreover, Loetscher et al. [20] similarly noted
individual's needs and the severity of the stroke. that the effectiveness of cognitive rehabilita-
A multi-disciplinary approach that involves close tion for attention deficits after stroke remains
collaboration between rehabilitation profession- unproven (20). In contrast, strong evidence has
als and the patient's family is often the most ef- been found for the effectiveness of speech thera-
fective.” py in post-stroke patients. According to Brady et
al. [21], speech and language therapy resulted in
Verification of responses to the Cochrane Data-
statistically and clinically significant benefits for
base's Systematic Reviews
functional communication, reading, writing, and
The Cochrane Library database was re-examined expressive language. Evidence was also found to
to identify and evaluate the efficacy of rehabilita- indicate the effectiveness of water‐based exer-
tion interventions. The study by Pollock et al. [18] cises for improving activities of daily living after
highlights the abundance of systematic reviews stroke [22]. Mehrholz et al. [22] found that, wa-
related to interventions for improving upper ter-based exercises after stroke showed a signif-
limb function post-stroke; however, the current icant improvement in activity of daily living and
evidence for these interventions is insufficient muscle strength; however, significant improve-
to make informed clinical decisions. A compre- ment was found in gait speed, postural balance,
hensive literature search was conducted, but no or fitness. Price et al. [23] revealed that while ES
updated publications were found in the database around the shoulder after a stroke may not influ-
resources related to the indicated 2014 paper. ence reports of pain, there do appear to be ben-
Concerning occupational therapy, a systemat- efits for passive humeral lateral rotation, possibly
ic review and meta-analysis by Gibson et al. [19] through the reduction of glenohumeral subluxa-
was found. The study revealed low-certainty ev- tion (Table 2).

Table 2. Verification of responses regarding neurological disorders.

ChatGPT Cochrane Database

Physical therapy Partially confirmed

Occupational therapy Confirmed

Speech therapy Confirmed

Cognitive rehabilitation Confirmed

Aquatic therapy Confirmed

Electrical stimulation Confirmed

10
 Physiotherapy Review | Volume XXVII Issue 4/2023

Chronic respiratory diseases uals with Idiopathic Pulmonary Fibrosis. Osadnik

et al. [25] conducted a SR to evaluate the effica-
ChatGPT question:
cy of pulmonary rehabilitation compared to the
“What are the effective rehabilitation treatments
standard of care for adult patients with asthma.
for chronic respiratory diseases?
The results suggest that pulmonary rehabilita-
ChatGPT answer: tion, compared to usual care, may increase max-
“Pulmonary rehabilitation is an effective rehabili- imal oxygen uptake and improve functional exer-
tation treatment for chronic respiratory diseases, cise capacity as measured by the 6-minute walk
which may include: distance with a moderate level of evidence.

• Aerobic exercise: to improve cardiovascular In contrast, Silva et al. [26] found no definitive
function and increase endurance evidence for or against the use of inspiratory
muscle exercises in the treatment of asthma. A
• Strength training: to improve muscle strength
significant proportion of the available systemat-
and support breathing
ic reviews regarding rehabilitation interventions
• Breathing techniques, such as pursed lip
has been examined on patients with Chronic Ob-
breathing and diaphragmatic breathing, to im-
structive Pulmonary Disease (COPD). The recent
prove breathing control and reduce shortness
evidence regarding the impact of exercise and
of breath
specific exercises aimed at improving breathing
• Education: to provide information and support muscle strength on breathlessness, physical fit-
for managing symptoms, medication use, and ness, and quality of life in patients with weakened
healthy lifestyle changes breathing muscles is inconclusive [27]. Malaguti
Nutritional counseling: to maintain a healthy et al. [28] reported evidence of low to moderate
weight and support respiratory function. certainty, suggesting that supervised mainte-
nance programs subsequent to pulmonary reha-
It is important to work with a rehabilitation spe-
bilitation may enhance health-related quality of
cialist to create a personalized rehabilitation plan
life between 6 to 12 months when compared to
that meets your needs and goals."
standard care. The findings regarding the effi-
Verification of responses to the Cochrane Data- cacy of educational interventions for healthcare
base's Systematic Reviews professionals in primary care to manage COPD
The resources of the Cochrane Library were also are uncertain. The interventions could potential-
searched in this area to investigate rehabilitation ly influence the proportion of COPD diagnoses
interventions. The analysis encompassed both confirmed using spirometry, patient involvement
obstructive and restrictive respiratory disorders. in pulmonary rehabilitation, and the utilization
The analysis of Dowman et al. [24] on intersti- of guideline-recommended COPD respiratory
tial lung disease revealed that both inpatient and medications [29]. However, among the studies re-
home-based rehabilitation interventions show viewed, a systematic review with meta-analysis
efficacy in enhancing exercise tolerance as eval- by Cox et al. reported on the effectiveness of tele-
uated using the 6-minute walk test. However, the rehabilitation interventions was identified. These
evidence regarding the improvement of dyspnea results were not included in the output from the
and quality of life in the short term is of low cer- ChatGPT tool (Table 3).
tainty, with similar benefits observed in individ-

11
Physiotherapy Review | Volume XXVII Issue 4/2023

Table 3. Verification of responses regarding chronic respiratory diseases.

ChatGPT Cochrane Database

Aerobic exercise Confirmed

Strength training Confirmed

Breathing techniques Confirmed

Education Confirmed

Nutritional counseling Confirmed

- Telerehabilitation

Discussion
This review aimed to evaluate the perfor- of "Cognitive-behavioral therapy" and "Heat and
mance and concordance of the newly developed cold therapy" compared to a control group with
ChatGPT tool by comparing its responses to the no intervention or a placebo intervention are
scientific literature on the efficacy of rehabilita- inconclusive. Therefore, the current ChatGPT’s
tion interventions for three of the most prevalent evidence base in this area appears limited. In
categories of functional disorders: musculoskele- the field of neurological disorders, the scientif-
tal, neurological, and chronic respiratory diseas- ic evidence found in Cochrane Library resources
es. In 2019, these three categories accounted for showed that the effective methods recommend-
a combined total of 2785 million cases. ed by the chatbot were consistent with the evi-
In terms of AI responses to treatments for chronic dence. However, it was noted that there is inad-
lower back pain, ChatGPT lists several commonly equate quality of evidence to support the use of
used treatments. A correspondence between the physical therapy for this purpose. Interestingly,
treatments listed by AI and the systematic re- the effectiveness of therapies for motor learning
views with meta-analyses found in the Cochrane skills after stroke after stroke revealed that the
Library database has been noted. The treatments neurophysiological rehabilitation methods that
listed by ChatGPT, such as exercise therapy, man- have been widely used for more than 80 years
ual therapy, spinal manipulation, acupuncture, lack "hard" scientific evidence of their effective-
and interventional procedures, are also evaluated ness. The AI's responses in the area of chronic
in the available scientific evidence regarding their respiratory diseases were found to be consistent
effectiveness in scientific databases. The AI's an- with established scientific evidence. Notably, the
swer generally aligns with the results found in chatbot did not mention telerehabilitation meth-
the Cochrane Library, with some differences in ods, which have garnered significant attention
the level of certainty of the evidence for reha- since the COVID-19 pandemic began. This ap-
bilitation interventions in stroke. Based on the proach has been the focus of recent research on
analysis of available meta-analyses, the benefits respiratory diseases.

12
 Physiotherapy Review | Volume XXVII Issue 4/2023

Although ChatGPT has shown a potential to iden- human medical writers as they lack the level of
tify effective rehabilitation methods for three of expertise and understanding in the field of medi-
the most frequent health problems, it would be cine. The use of language models in medical writ-
necessary to reflect now if ChatGPT could make ing also raises ethical concerns, and it is impor-
a recommendation, not only about the method tant for organizations and individuals to carefully
but also of identifying the dose, intensity, and consider these issues to ensure the quality and
frequency of the interventions to be done for accuracy of the generated documents. The paper
the patients. In this sense, for example, physical by Shen et al. [31] describes the potential use of
exercise as an effective intervention against low AI-based chatbots in the healthcare industry. The
back pain presents in the scientific literature a authors highlight the use of ChatGPT in improv-
high variability in the different possibilities of ex- ing the accuracy of radiology imaging exams. The
ercises (volume and intensities) that are effective authors emphasized the significance of utilizing
depending on the clinical moment, patient, and the capabilities of this technology to facilitate
process. At this point, it would be worth reflect- clinical decision-making processes, as well as for
ing on whether ChatGPT will be able to refine its evaluating the accuracy of imaging techniques.
response to indicate, among the different meth- The study conducted by Huh compared the par-
ods identified in the present review, whether it asitology knowledge and interpretation capabil-
could analyze which of them may have a better ities of ChatGPT to that of medical students in
response, thus identifying classification algo- Korea [32]. The results showed that AI performed
rithms for complex clinical decision-making. lower than medical students, and its correct an-
swer rate was not related to the difficulty level of
It is challenging to predict the future utilization
the items. The study concluded that ChatGPT's
of AI in scientific papers. Nonetheless, there have
knowledge and interpretation ability in the field
been instances of experimentation with AI in sci-
of parasitology is different from that of medical
entific writing, specifically regarding text gener-
students in Korea. The study by O'Connor rais-
ation and proofreading. The use of ChatGPT was
es concerns about the fact that it may soon be-
tested by researchers at Northwestern University
come difficult to distinguish between a student's
in Chicago (USA), where the tool generated ab-
labor and that of an AI chatbot, especially at the
stracts of scientific publications. Results showed
post-graduate level, and calls for a re-evalua-
that in a survey, one-third of participants were
tion of the types of assessment used in nursing
unable to differentiate between text generated by
education [33]. The last publication in Stem Cell
ChatGPT and text written by a human. A review
Reports presented an interview with the guest
of PubMed on February 6, 2023, revealed twen-
editors, Cahan and Treutlein [34], who discussed
ty-one articles related to the use of AI in scien-
their utilization of AI for computational systems
tific writing, with only five related to the field of
biology. They highlighted ChatGPT's capability to
medicine. In a paper by Biswas, the potential use
effectively handle vast amounts of data, identify
of ChatGPT in medical writing was explored [30].
patterns and trends, optimize stem cell culture
The paper argues that chatbots and NLP technol-
conditions, and construct comprehensive models
ogy, such as GPT, could enhance the efficiency of
to simulate stem cell behavior.
the medical writing process by automating cer-
tain tasks and improving the accuracy and con- Remarkably, ChatGPT consistently advises users
sistency of documents. For example, chatbots to consult with a rehabilitation specialist, pro-
could extract information from electronic med- viding a valuable safeguard for practitioners. This
ical records, generate draft text, and assist with tool may serve as a helpful resource for individu-
literature searches. However, the paper acknowl- als with limited medical knowledge, particularly
edges that ChatGPT can only partially replace in countries where rehabilitation services within

13
Physiotherapy Review | Volume XXVII Issue 4/2023

the healthcare system are limited. Raising aware- Funding:

ness among patients of the various therapeutic This research received no external funding.
options available for a given dysfunction or dis-
order could increase the range of interventions Acknowledgments:
they receive. The effectiveness of ChatGPT as a The author acknowledges this article was sup-
diagnostic tool remains uncertain, but it may of- ported by ChatGPT (powered by OpenAI's lan-
fer a more reliable solution than "Dr. Google." guage model, GPT-3; http://openai.com). The
author would like to thank Radek Kobialko for
Conclusions sparking interest in the topic.
Based on the analysis, it can be concluded that
Competing interests:
using a freely available AI language model, such
The authors declare no conflict of interest.
as ChatGPT, has the potential to assist in iden-
tifying effective rehabilitation methods for com-
mon conditions. The analysis of one-sentence
questions on therapeutic methods across all are-
as indicated the presence of established scientific
methods, although not all available rehabilitation
methods were demonstrated in certain areas.
The AI's responses were generally consistent
with the evidence found in the Cochrane Library,
with some differences in the level of certainty of
the evidence. In the area of chronic lower back
pain, the AI listed commonly used treatments
that align with the available scientific evidence.
At the same time, the benefits of CBT and heat
and cold therapy are inconclusive. The scientif-
ic evidence found in Cochrane Library resources
showed that the effective methods recommend-
ed by the chatbot were consistent with the evi-
dence in neurological disorders. However, there
is inadequate quality of evidence to support the
use of physical therapy for this purpose. The AI's
responses in the area of chronic respiratory dis-
eases were consistent with established scientific
evidence, but it did not mention telerehabilitation
methods, which have been the focus of recent re-
search in this area. The results of this study pro-
vide valuable insights into the potential applica-
tions of AI in the field of rehabilitation and may
serve as a basis for future research in this area.

14
 Physiotherapy Review | Volume XXVII Issue 4/2023

References
1. Pajak P. It has only been on the market for 2 months, 11. Hayden JA, Ellis J, Ogilvie R, Malmivaara A, van
and already it is wreaking massive havoc. 2023. Tulder MW. Exercise therapy for chronic low back
2. Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones pain. Cochrane Database Syst Rev. 2021; 9 (9):
L, Gomez AN, et al. Attention is all you need. Ad- CD009790.
vances in neural information processing systems. 12. Mu J, Furlan AD, Lam WY, Hsu MY, Ning Z, Lao L.
2017;30 (Part of Advances in Neural Information Acupuncture for chronic non-specific low back
Processing Systems 30 – NIPS 2017). pain. Cochrane Database Syst Rev. 2020; 12 (12):
3. What Is the Transformer Architecture and How CD013814.
Does It Work? 2023. Available at: https://datagen. 13. Rubinstein SM, van Middelkoop M, Assendelft WJ,
tech/guides/computer-vision/transformer-archi- de Boer MR, van Tulder MW. Spinal manipulative
tecture/ therapy for chronic low-back pain. Cochrane Data-
4. Vanian J. Why tech insiders are so excited about base Syst Rev. 2011; (2): CD008112.
ChatGPT, a chatbot that answers questions and 14. Wieland LS, Skoetz N, Pilkington K, Harbin S, Ve-
writes essays. 2022. Available at: https://www. mpati R, Berman BM. Yoga for chronic non-specific
linkedin.com/pulse/why-tech-insiders-so-excit- low back pain. Cochrane Database Syst Rev. 2022; 11
ed-chatgpt-chatbot-answers-questions-garg (11): CD010671.
5. Montti R. What is ChatGPT And How Can You Use 15. Williams ACC, Fisher E, Hearn L, Eccleston C. Psy-
It? 2022. Available at: https://www.searchengine- chological therapies for the management of chron-
journal.com/what-is-chatgpt/473664/ ic pain (excluding headache) in adults. Cochrane
6. Elsen-Rooney M. NYC education depart- Database Syst Rev. 2020; 8 (8): CD007407.
ment blocks ChatGPT on school devices, net- 16. Schaafsma FG, Whelan K, van der Beek AJ, van der
works. 2023. Available at: https://www.chalk- Es-Lambeek LC, Ojajärvi A, Verbeek JH. Physical
b e a t . o r g / n e w y o r k / 2 0 2 3/ 1 / 3/ 2 3 5 3 7 9 8 7/ conditioning as part of a return to work strategy
nyc-schools-ban-chatgpt-writing-artificial-intelli- to reduce sickness absence for workers with back
gence/ pain. Cochrane Database Syst Rev. 2013; 2013 (8):
7. Ault A. AI Bot ChatGPT Passes US Medical Licens- Cd001822.
ing Exams Without Cramming – Unlike Students. 17. French SD, Cameron M, Walker BF, Reggars JW,
2023. Available at: https://healthitanalytics.com/ Esterman AJ. Superficial heat or cold for low back
news/chatgpt-passes-us-medical-licensing-ex- pain. Cochrane Database Syst Rev. 2006; 2006 (1):
am-without-clinician-input CD004750.
8. Stokel-Walker C. ChatGPT listed as author on re- 18. Pollock A, Farmer SE, Brady MC, Langhorne P, Mead
search papers: many scientists disapprove. Nature. GE, Mehrholz J, et al. Interventions for improving
2023; 613 (7945): 620–621. upper limb function after stroke. Cochrane Data-
9. Howick J, Koletsi D, Ioannidis JPA, Madigan C, Pan- base Syst Rev. 2014; 2014 (11): CD010820.
dis N, Loef M, et al. Most healthcare interventions 19. Gibson E, Koh CL, Eames S, Bennett S, Scott AM,
tested in Cochrane Reviews are not effective ac- Hoffmann TC. Occupational therapy for cognitive
cording to high quality evidence: a systematic re- impairment in stroke patients. Cochrane Database
view and meta-analysis. J Clin Epidemiol. 2022; 148: Syst Rev. 2022; 3 (3): CD006430.
160–169. 20. Loetscher T, Potter KJ, Wong D, das Nair R. Cog-
10. Cieza A, Causey K, Kamenov K, Hanson SW, Chat- nitive rehabilitation for attention deficits following
terji S, Vos T. Global estimates of the need for re- stroke. Cochrane Database Syst Rev. 2019; 2019 (11):
habilitation based on the Global Burden of Disease CD002842.
study 2019: a systematic analysis for the Global
Burden of Disease Study 2019. Lancet. 2021; 396
(10267): 2006–2017.

15
 Physiotherapy Review | Volume XXVII Issue 4/2023

21. Brady MC, Kelly H, Godwin J, Enderby P, Campbell 28. Malaguti C, Dal Corso S, Janjua S, Holland AE. Su-
P. Speech and language therapy for aphasia follow- pervised maintenance programmes following pul-
ing stroke. Cochrane Database Syst Rev. 2016; 2016 monary rehabilitation compared to usual care for
(6): CD000425. chronic obstructive pulmonary disease. Cochrane
22. Mehrholz J, Kugler J, Pohl M. Water-based exer- Database Syst Rev. 2021; 8 (8): CD013569.
cises for improving activities of daily living after 29. Cross AJ, Thomas D, Liang J, Abramson MJ, George
stroke. Cochrane Database Syst Rev. 2011; 2011 (1): J, Zairina E. Educational interventions for health
CD008186. professionals managing chronic obstructive pul-
23. Price CI, Pandyan AD. Electrical stimulation for monary disease in primary care. Cochrane Data-
preventing and treating post-stroke shoulder base Syst Rev. 2022; 5 (5): CD012652.
pain. Cochrane Database Syst Rev. 2000; 2000 (4): 30. Biswas S. ChatGPT and the Future of Medical Writ-
CD001698. ing. Radiology. 2023; 307 (2): e223312.
24. Dowman L, Hill CJ, May A, Holland AE. Pulmonary 31. Shen Y, Heacock L, Elias J, Hentel KD, Reig B, Shih
rehabilitation for interstitial lung disease. Cochrane G, et al. ChatGPT and Other Large Language Mod-
Database Syst Rev. 2021 ; 2 (2): CD006322. els Are Double-edged Swords. Radiology. 2023: 307
25. Osadnik CR, Gleeson C, McDonald VM, Holland (2): e230163.
AE. Pulmonary rehabilitation versus usual care for 32. Huh S. Are ChatGPT's knowledge and interpreta-
adults with asthma. Cochrane Database Syst Rev. tion ability comparable to those of medical stu-
2022; 8 (8): CD013485. dents in Korea for taking a parasitology examina-
26. Silva IS, Fregonezi GA, Dias FA, Ribeiro CT, Guer- tion?: a descriptive study. J Educ Eval Health Prof.
ra RO, Ferreira GM. Inspiratory muscle training for 2023; 20: 1.
asthma. Cochrane Database Syst Rev. 2013; 2013 (9): 33. O'Connor S, ChatGpt. Open artificial intelligence
CD003792. platforms in nursing education: Tools for academ-
27. Ammous O, Feki W, Lotfi T, Khamis AM, Gosselink ic progress or abuse? Nurse Educ Pract. 2023; 66:
R, Rebai A, et al. Inspiratory muscle training, with 103537.
or without concomitant pulmonary rehabilitation, 34. Cahan P, Treutlein B. A conversation with ChatGPT
for chronic obstructive pulmonary disease (COPD). on the role of computational systems biology in stem
Cochrane Database Syst Rev. 2023;1 (1): CD013778. cell research. Stem Cell Reports. 2023; 18 (1): 1–2.

16

View publication stats